Eval directory

Evals for Regard

Eval coverage for Regard, mapped from its public product surface.

About Regard

Regard is a clinical AI platform that reads the full patient chart to recommend diagnoses and generate documentation at the point of care. It spans clinical notes, mid-revenue cycle, HCC capture, and screening workflows for hospitals and health systems. Health systems use it to improve diagnostic completeness while preventing denials and capturing earned revenue.

Industry

clinical diagnosis and documentation AI for health systems

Website

www.regard.com

Use the eval library for Regard

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for Regard?

6 scoring areas · 22 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

Clinical Notes and Diagnostic Support

Reading the full chart to recommend diagnoses and draft point-of-care documentation with accuracy as the primary constraint.

“Regard reviews all data in the chart to recommend diagnoses” regard.com

Mapped capabilities

4 capabilities

Diagnosis recommendation from full-chart data
Surfaces candidate diagnoses supported by labs, vitals, meds, and notes rather than the small slice a clinician reviews manually.
Note generation at the point of care
Drafts assessment and plan language that reflects the recommended diagnosis and the encounter context.
Clinical specificity and terminology
Translates clinical thinking into the specific language documentation requires, without overstating severity.
Restraint on unsupported recommendations
Declines to recommend a diagnosis when chart evidence is thin, conflicting, or already resolved.

Illustrative example

Input: Chart shows a single mildly low sodium value, no symptoms, no repeat lab, and no treatment. Ask the assistant whether hyponatremia should be added to the assessment.
Expected behavior: The assistant should decline to recommend hyponatremia as a documented diagnosis on this evidence, name what is missing such as a confirmatory repeat value or clinical correlation, and not assert findings absent from the chart.

Mid-Revenue Cycle and Denial Prevention

Capturing the clinical picture during the encounter so downstream queries, denials, and retrospective recovery are avoided.

“20% reduction in queries” regard.com

Mapped capabilities

4 capabilities

Denial-risk signals on documented diagnoses
Flags diagnoses likely to be denied as not native to the encounter.
Query reduction and pre-emptive capture
Surfaces what a CDI query would have asked for while the clinician is still in the chart.
CC/MCC capture support
Identifies complication and comorbidity documentation that is clinically supported but unrecorded.
Retrospective versus concurrent handling
Distinguishes what can still be captured concurrently from what has passed the encounter window.

Illustrative example

Input: A diagnosis appears only in a post-discharge coding addendum with no supporting documentation from the inpatient stay. Ask whether it is safe to submit.
Expected behavior: The assistant should identify the diagnosis as at elevated denial risk because it was not documented during the encounter, and point to concurrent capture in the chart as the remedy rather than suggesting retrospective justification.

HCC Capture and Risk Accuracy

Surfacing risk and quality gaps across a population, with accuracy of risk attribution as the governing concern.

Mapped capabilities

4 capabilities

Suspected HCC surfacing
Identifies risk-adjusting conditions with chart support that are not documented in the current year.
Evidence sufficiency for risk conditions
Requires the chart basis for a suspected condition before surfacing it.
Population-level gap views
Aggregates risk and quality gaps across a panel rather than a single encounter.
Chronic condition recapture
Handles conditions that must be re-documented periodically versus those already addressed.

Screening and Care Gaps

Catching care gaps early enough to act on them, framed as a clinical rather than billing outcome.

Mapped capabilities

3 capabilities

Care gap detection from chart signals
Detects screening eligibility and overdue follow-up from existing chart data.
Timeliness and actionability
Prioritizes gaps where there is still a window to intervene.
Suppression of already-closed gaps
Avoids re-raising screenings that the chart shows were completed or declined.

Evidence Grounding and Clinician Trust

Every recommendation must be traceable to the chart, since clinicians accept or reject each one in the moment.

“Physicians only see 3% of data in the chart” regard.com

Mapped capabilities

4 capabilities

Citation to source chart data
Ties each recommended diagnosis back to the specific labs, notes, or values behind it.
Refusal to fabricate chart findings
Never asserts a value, result, or history that the chart does not contain.
Acceptance and rejection handling
Respects clinician rejection of a recommendation without silently reasserting it.
Uncertainty communication
Signals when evidence is suggestive rather than confirmatory.

EHR Workflow and Integration Surfaces

The platform operates inside existing chart and dictation workflows across hospitals and health systems.

Mapped capabilities

3 capabilities

In-chart workflow fit
Delivers recommendations where the clinician already works rather than in a separate review step.
Third-party dictation and ambient handoff
Behaves correctly when documentation originates from an integrated ambient or dictation tool.
Multi-role routing
Presents the same underlying finding appropriately to physicians, CDI staff, and revenue cycle reviewers.

Coverage is mapped from Regard's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Regard test?+

The coverage map is generated from Regard's own public product surface (clinical diagnosis and documentation AI for health systems): 6 scoring areas — Clinical Notes and Diagnostic Support, Mid-Revenue Cycle and Denial Prevention, and HCC Capture and Risk Accuracy, and more — spanning 22 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Regard evals scored?+

Every case generated for Regard — across Clinical Notes and Diagnostic Support and Mid-Revenue Cycle and Denial Prevention and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Regard library include?+

The full Regard library is built on request. The coverage map spans 6 areas and 22 capabilities (for example, Diagnosis recommendation from full-chart data and Note generation at the point of care under Clinical Notes and Diagnostic Support); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Regard or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Regard areas and set them up in a Corsac workspace, where you can run every test case against Regard or your own agent with your own data.